Categorization
Participants assign labels from your custom taxonomy to each item. You get label distributions, confidence scores, and inter-annotator agreement — without building any labeling infrastructure.
Candor does this so you don't have to
Define your label set
Provide any set of labels that match your classification needs. Safe/borderline/violation, positive/neutral/negative, or fully custom categories. Candor builds the participant interface automatically.
20 items per batch
Items are grouped into batches of 20 for fast classification. Once participants learn the taxonomy, they stay in flow and produce consistent labels with minimal fatigue.
Multiple annotators label same items
Each item is labeled by multiple participants — not just one. Overlapping annotations let Candor measure inter-annotator agreement and produce reliable label distributions.
Distributions and confidence computed
Label distributions and confidence scores are computed automatically from all annotations. You see the full picture — not just a majority vote — with disagreement clearly surfaced.
{ confidence: 0.85 }Content Moderation
Classify user-generated content as safe, borderline, or violation at scale with reliable inter-annotator agreement.
Sentiment Analysis
Label text as positive, neutral, or negative. Get full sentiment distributions for nuanced content.
Image Classification
Categorize images into custom taxonomies. Multiple annotators ensure labeling quality and consistency.
Hallucination Detection
Label LLM outputs as factual, uncertain, or hallucinated. Confidence scores reveal where models struggle.
Red Teaming Safety
Classify adversarial model outputs by severity. Multi-annotator overlap catches edge cases a single reviewer would miss.
Document Categorization
Sort documents into topic categories, compliance levels, or priority tiers with high-throughput batching.